# コンピュータ大貧民における高速な相手モデル作成と精度向上The Accuracy Improvement with The fast Opponent Modeling in The Computer DAIHINMIN

## 抄録

UEC コンピュータ大貧民大会ではモンテカルロ法を用いたクライアントが優勝している．そこでプレイアウト中の相手着手を実際の着手に近づけることでモンテカルロ法によるクライアントの強化を考える．本研究ではゲーム中の実際の相手着手を学習する方法としてナイーブベイズを用いる．これにより高速な相手のモデル化を行う．さらに、学習素性の工夫により精度の向上を行った．この結果，作成されたモデルの精度は過去の優勝クライアント snowl に対し，4 割程度の近似ができた．Monte-Carlo method is also useful for DAIHNMIN and the client using this method has won the UEC computer DAIHINMIN tournament. We try to accelerate the strength of Monte-Carlo method by making effective opponent models which are close to the real opponents' moves. Stronger opponent models, more effective playouts our client has. We use Naive Bayes as the learning method to modeling the opponents. This method is one of the fastest algorithm for learning and classification. In addition, its accuracy is enough to modeling the opponents. In this paper, we show two modeling by Naive Bayes. The first method is the simple modeling, and the second is improved the move data structure. The accuracy is approximately 40% by our improved method to model snowl which is the champion client in 2010.

Monte-Carlo method is also useful for DAIHNMIN and the client using this method has won the UEC computer DAIHINMIN tournament. We try to accelerate the strength of Monte-Carlo method by making effective opponent models which are close to the real opponents' moves. Stronger opponent models, more effective playouts our client has. We use Naive Bayes as the learning method to modeling the opponents. This method is one of the fastest algorithm for learning and classification. In addition, its accuracy is enough to modeling the opponents. In this paper, we show two modeling by Naive Bayes. The first method is the simple modeling, and the second is improved the move data structure. The accuracy is approximately 40% by our improved method to model snowl which is the champion client in 2010.

## 収録刊行物

• 研究報告数理モデル化と問題解決（MPS）

研究報告数理モデル化と問題解決（MPS） 2013-MPS-96(4), 1-3, 2013-12-04

一般社団法人情報処理学会

## 各種コード

• NII論文ID(NAID)
110009634052
• NII書誌ID(NCID)
AN10505667
• 本文言語コード
JPN
• 資料種別
Technical Report
• ISSN
09196072
• データ提供元
NII-ELS  IPSJ

ページトップへ